CoreViz
SDK

Vision SDK Overview

AI-powered image analysis and manipulation with @coreviz/sdk

Vision SDK

The @coreviz/sdk npm package gives you access to CoreViz's AI vision capabilities — describing, tagging, editing, generating, and embedding images — directly from your JavaScript, TypeScript, or React Native application.

Installation

npm install @coreviz/sdk

React Native / Expo

npx expo install expo-image-manipulator expo-file-system

React Native limitation

Local mode (mode: 'local') for tag() and embed() is not supported on React Native / Expo.

Setup

import { CoreViz } from '@coreviz/sdk';

const sdk = new CoreViz({
  apiKey: process.env.COREVIZ_API_KEY
});

Available Methods

describe()

Generate a detailed text description or caption for any image URL or base64 string.

Learn More

tag()

Classify images with custom prompts and option lists. Supports both API and local in-browser processing.

Learn More

embed() + similarity()

Generate CLIP embeddings for images or text, then compute cosine similarity for semantic search.

Learn More

edit() + generate()

Edit images with natural language prompts, or generate new images from text descriptions.

Learn More

Quick Start

import { CoreViz } from '@coreviz/sdk';

const sdk = new CoreViz({ apiKey: process.env.COREVIZ_API_KEY });

// Describe an image
const description = await sdk.describe('https://example.com/photo.jpg');
console.log(description); // "A red sports car parked on a cobblestone street..."

// Tag / classify
const { tags } = await sdk.tag('https://example.com/photo.jpg', {
  prompt: 'What type of vehicle is this?',
  options: ['car', 'truck', 'motorcycle', 'bus'],
  multiple: false,
});
console.log(tags); // ["car"]

// Generate embeddings for semantic search
const { embedding } = await sdk.embed('https://example.com/photo.jpg', { type: 'image' });
const { embedding: queryEmbed } = await sdk.embed('red sports car', { type: 'text' });
const score = sdk.similarity(embedding, queryEmbed);
console.log(score); // 0.82

// Edit an image
const editedUrl = await sdk.edit('https://example.com/photo.jpg', {
  prompt: 'Make it look like a watercolor painting',
  aspectRatio: '1:1',
});

Authentication

All methods require an API key. Get yours from Your Account → API Keys on lab.coreviz.io.

// Via constructor
const sdk = new CoreViz({ apiKey: 'your_api_key' });

// Or for users authenticated via coreviz login
const sdk = new CoreViz({ token: 'session_token' });

Supported Image Inputs

All vision methods accept:

  • URL — publicly accessible HTTPS URL
  • Base64 — data URI string (for local files)
// URL
await sdk.describe('https://cdn.example.com/photo.jpg');

// Base64
const base64 = await fileToBase64(file);
await sdk.describe(base64);

// Auto-resize before sending (saves bandwidth)
const resized = await sdk.resize(file, 1920, 1080);
await sdk.describe(resized);

resize()

Resize images client-side or server-side before processing to reduce bandwidth and improve speed.

const resized = await sdk.resize(imageFileOrUrl, 800, 600);
// or as a standalone import:
import { resize } from '@coreviz/sdk';

Also available as a standalone export for use without creating a CoreViz instance.

Raw API Reference

If you prefer to call the HTTP endpoints directly, see the Vision API Reference.